Skip to content
Stan Seibert edited this page Oct 30, 2018 · 1 revision

Numba Meeting: 2018-10-30

Attendees: Stuart, Ehsan, Todd, Stan, Siu

0. Feature Discussion

  • LLVM 7 status
    • breaking with SVML patch
    • llvmlite builds+tests fine
    • numba tests fine on linux; issues on win/osx
    • moving forward:
      • temp adjust version string for llvm7 builds to this the stack
  • String status
    • PR overload(operator.getitem) #3375 updated
    • Added perf patch to disable refcount ops in critical loop
    • Unable to vectorize due to branch (see comment)
      • manual partial unroll?
    • move performance enhancement to another PR
  • Debug logging for caught exception

1. New issues

  • 3453 - Report actual problem in known places of constant inference failure
    • request for better error msg
  • 3451 - Nested Jitclass Generator Memory Leak
    • confirmed bug
  • 3448 - Link to jit-decorator-fastmath missing
    • fixed with PR
  • 3446 - problem has occurred in Numba's internals.
  • 3445 - Issue when trying to use python's any
    • known generator issue
  • 3443 - Inject traceback in cuda raised exception
    • feature request
  • 3442 - Implement NumPy trapz
  • 3441 - funcsigs/inspect signature support (PEP 362 function signature object)
  • 3439 - NotImplementedError followed by LoweringError
  • 3438 - Add (or document) environment variable to disable parallelism

2. Open PRs

New

  • 3452 - Fix broken link and mark up problem.
    • can merge when buildfarm is ready
  • 3450 - [WIP] generated_jit for CUDA kernels
  • 3449 - [WIP] Allow matching non-array objects in find_callname()
  • 3437 - Changes to accommodate LLVM 7.0.x

Old

  • 3429 - Fixes raising of user defined exceptions for exec(<string>).
  • 3416 - Optimized usage of alloca out of the loop
  • 3414 - [WIP] Refactor Const type
  • 3403 - Fix label offset in call inline after parfor pass
  • 3402 - Support negative indexing in tuple
  • 3399 - Add max_registers Option to cuda.jit
  • 3397 - Fix with-objmode warning
  • 3392 - Launch and attach gdb directly from Numba.
  • 3390 - typeinfer: use unknown_loc object instead of string literal
  • 3385 - conda recipe: whitelist libiomp5.dylib
  • 3382 - CUDA_ERROR_MISALIGNED_ADDRESS Using Multiple Const Arrays
  • 3375 - Enable overload(operator.getitem)
  • 3363 - [WIP] string support
    • see above discussion
  • 3345 - Support for np.cov
    • ready for merge
  • 3162 Support constant dtype string in nopython mode in functions like numpy.empty.
    • Need to resolve #3195
  • 3160 First attempt at parallel diagnostics
    • Stuart will implement Todd's suggestion
  • 3134 [WIP] Cfunc x86 abi
    • Needs re-review
  • 3046 Pairwise sum implementation.
  • #2999 Support LowLevelCallable
  • #2950 Fix dispatcher to only consider contiguous-ness.
  • #2942 Fix linkage nature (declspec(dllexport)) of some test functions
  • #2894: [WIP] Implement jitclass default constructor arguments.
  • #2817: [WIP] Emit LLVM optimization remarks

===========================

4. Next Release: Version 0.41, RC=Nov 19, Final=Nov 26, 2018

  • Type refactoring
  • Parallel diagnostics
  • LLVM 7
  • Initial string support
  • Finishing off stalled PRs
  • Usual collection of bug fixes and small features
Clone this wiki locally